{"title":"基于机器学习的HL-2A设备边界特征量多输出预测","authors":"Zelong Li, Peng Yu, Qianhong Huang, Qi Zeng, Qingyi Tan, Yijun Zhong, Zhe Wang, Haoran Ye, Zhanhui Wang, Wulv Zhong, Min Xu","doi":"10.1007/s10894-025-00499-y","DOIUrl":null,"url":null,"abstract":"<div><p>The study of heat flux and particle transport in the plasma boundary and divertor region is a key issue for the long-term stable operation of the fusion reactor in the future. SOLPS-ITER is one of the most widely used boundary simulation programs, however, its calculation cost is high, and the calculation time is long. To enable the effective and rapid prediction of characteristic quantities in the DSOL region and meet the physical coupling requirements between the boundary and core regions (DSOL region and plasma core), integrated simulation for fast core-edge coupling is necessary. By using the SOLPS-ITER code and combining the parameters of the HL-2A device, the influence of impurity injection on the physical characteristics of the divertor boundary is studied, and the relevant simulation data are obtained. Two reliable prediction models of plasma boundary feature quantities are constructed, which are fully connected neural network model (DSOL-NN) and convolutional neural network model (DSOL-CNN). In order to better meet the needs of fast integrated simulation of plasma core-edge coupling, a multi-input multi-output mode (MIMO) is adopted. The model considers the effects of different impurity species and injection rates on the electron temperature and particle flux density of the divertor target plate. The results show that both models can successfully predict the electron temperature of the divertor target plate, the particle flux density of the target plate and the core-edge <b><i>Z</i></b><sub><b><i>eff</i></b></sub> under different impurity injection rate conditions. In comparison, the convolutional neural network model in the two models shows better prediction performance, with a mean relative error of about 5%, which is less than 10% of the fully connected neural network. A large number of comparative predictions show that the neural network prediction model takes several orders of magnitude less than the SOLPS-ITER simulation time consuming, thus providing a basis for the rapid integrated simulation of core-edge coupling.</p></div>","PeriodicalId":634,"journal":{"name":"Journal of Fusion Energy","volume":"44 1","pages":""},"PeriodicalIF":1.9000,"publicationDate":"2025-05-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Multi-Output Prediction of HL-2A Device Boundary Characteristic Quantities Based on Machine Learning\",\"authors\":\"Zelong Li, Peng Yu, Qianhong Huang, Qi Zeng, Qingyi Tan, Yijun Zhong, Zhe Wang, Haoran Ye, Zhanhui Wang, Wulv Zhong, Min Xu\",\"doi\":\"10.1007/s10894-025-00499-y\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><p>The study of heat flux and particle transport in the plasma boundary and divertor region is a key issue for the long-term stable operation of the fusion reactor in the future. SOLPS-ITER is one of the most widely used boundary simulation programs, however, its calculation cost is high, and the calculation time is long. To enable the effective and rapid prediction of characteristic quantities in the DSOL region and meet the physical coupling requirements between the boundary and core regions (DSOL region and plasma core), integrated simulation for fast core-edge coupling is necessary. By using the SOLPS-ITER code and combining the parameters of the HL-2A device, the influence of impurity injection on the physical characteristics of the divertor boundary is studied, and the relevant simulation data are obtained. Two reliable prediction models of plasma boundary feature quantities are constructed, which are fully connected neural network model (DSOL-NN) and convolutional neural network model (DSOL-CNN). In order to better meet the needs of fast integrated simulation of plasma core-edge coupling, a multi-input multi-output mode (MIMO) is adopted. The model considers the effects of different impurity species and injection rates on the electron temperature and particle flux density of the divertor target plate. The results show that both models can successfully predict the electron temperature of the divertor target plate, the particle flux density of the target plate and the core-edge <b><i>Z</i></b><sub><b><i>eff</i></b></sub> under different impurity injection rate conditions. In comparison, the convolutional neural network model in the two models shows better prediction performance, with a mean relative error of about 5%, which is less than 10% of the fully connected neural network. A large number of comparative predictions show that the neural network prediction model takes several orders of magnitude less than the SOLPS-ITER simulation time consuming, thus providing a basis for the rapid integrated simulation of core-edge coupling.</p></div>\",\"PeriodicalId\":634,\"journal\":{\"name\":\"Journal of Fusion Energy\",\"volume\":\"44 1\",\"pages\":\"\"},\"PeriodicalIF\":1.9000,\"publicationDate\":\"2025-05-17\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Fusion Energy\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://link.springer.com/article/10.1007/s10894-025-00499-y\",\"RegionNum\":4,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"NUCLEAR SCIENCE & TECHNOLOGY\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Fusion Energy","FirstCategoryId":"5","ListUrlMain":"https://link.springer.com/article/10.1007/s10894-025-00499-y","RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"NUCLEAR SCIENCE & TECHNOLOGY","Score":null,"Total":0}
Multi-Output Prediction of HL-2A Device Boundary Characteristic Quantities Based on Machine Learning
The study of heat flux and particle transport in the plasma boundary and divertor region is a key issue for the long-term stable operation of the fusion reactor in the future. SOLPS-ITER is one of the most widely used boundary simulation programs, however, its calculation cost is high, and the calculation time is long. To enable the effective and rapid prediction of characteristic quantities in the DSOL region and meet the physical coupling requirements between the boundary and core regions (DSOL region and plasma core), integrated simulation for fast core-edge coupling is necessary. By using the SOLPS-ITER code and combining the parameters of the HL-2A device, the influence of impurity injection on the physical characteristics of the divertor boundary is studied, and the relevant simulation data are obtained. Two reliable prediction models of plasma boundary feature quantities are constructed, which are fully connected neural network model (DSOL-NN) and convolutional neural network model (DSOL-CNN). In order to better meet the needs of fast integrated simulation of plasma core-edge coupling, a multi-input multi-output mode (MIMO) is adopted. The model considers the effects of different impurity species and injection rates on the electron temperature and particle flux density of the divertor target plate. The results show that both models can successfully predict the electron temperature of the divertor target plate, the particle flux density of the target plate and the core-edge Zeff under different impurity injection rate conditions. In comparison, the convolutional neural network model in the two models shows better prediction performance, with a mean relative error of about 5%, which is less than 10% of the fully connected neural network. A large number of comparative predictions show that the neural network prediction model takes several orders of magnitude less than the SOLPS-ITER simulation time consuming, thus providing a basis for the rapid integrated simulation of core-edge coupling.
期刊介绍:
The Journal of Fusion Energy features original research contributions and review papers examining and the development and enhancing the knowledge base of thermonuclear fusion as a potential power source. It is designed to serve as a journal of record for the publication of original research results in fundamental and applied physics, applied science and technological development. The journal publishes qualified papers based on peer reviews.
This journal also provides a forum for discussing broader policies and strategies that have played, and will continue to play, a crucial role in fusion programs. In keeping with this theme, readers will find articles covering an array of important matters concerning strategy and program direction.